Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 76051.66331605377,
-      "required_pages_per_second_upper_estimate": 0.006591965425057083
+      "required_token_positions_per_second": 74734.8864154142,
+      "required_pages_per_second_upper_estimate": 0.006477830540650189
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9506.457914506722,
-      "required_pages_per_second_upper_estimate": 0.003792268196308729
+      "required_token_positions_per_second": 9341.860801926776,
+      "required_pages_per_second_upper_estimate": 0.00372660794715445
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 19012.915829013444,
-      "required_pages_per_second_upper_estimate": 0.0043983034611522216
+      "required_token_positions_per_second": 18683.72160385355,
+      "required_pages_per_second_upper_estimate": 0.004322150170781966
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28519.373743520166,
-      "required_pages_per_second_upper_estimate": 0.0017366802313903299
+      "required_token_positions_per_second": 28025.582405780326,
+      "required_pages_per_second_upper_estimate": 0.001706610929644901
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28519.373743520166,
-      "required_pages_per_second_upper_estimate": 0.0021955003374555166
+      "required_token_positions_per_second": 28025.582405780326,
+      "required_pages_per_second_upper_estimate": 0.0021574869133743953
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.042780299000014566,
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
-      "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28519.373743520166,
-      "required_pages_per_second_upper_estimate": 0.3541151736906039
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 28025.582405780326,
+      "required_pages_per_second_upper_estimate": 0.347983937889173
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 62,
+    "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.23414474007012132,
-    "producer_p95_seconds": 0.2175090969994926,
-    "gpu_update_seconds": 0.9694462525244564,
-    "producer_batches_per_second": 4.270862542974578,
-    "gpu_batches_per_second": 1.03151670079283,
-    "producer_headroom": 4.140371687333775,
+    "producer_batch_seconds": 0.05783258674225067,
+    "producer_p95_seconds": 0.20834068400108663,
+    "gpu_update_seconds": 0.9865272235804652,
+    "producer_batches_per_second": 17.291289501830832,
+    "gpu_batches_per_second": 1.0136567710424018,
+    "producer_headroom": 17.05832782436722,
     "sustainable": true,
-    "buffer_seconds": 60.105667656516296,
+    "buffer_seconds": 60.17816063840838,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 75,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 377120.70832940046,
-    "upload_bytes_per_second": 6032431.446804246,
+    "generation_bytes_per_second": 397308.96389371104,
+    "upload_bytes_per_second": 6119065.179176733,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 80.9213250867034,
+    "estimated_full_shards_per_hour": 85.25325477226733,
     "selection_policy": "Smallest shard within 90% of measured local peak, enlarged to amortize 30 seconds of measured page production; bounded to 256 MiB."
   },
   "upload_measurements": [
-    {
-      "seconds": 1.0632401930001834,
-      "bytes": 22792,
-      "pages": 1,
-      "shards": 1
-    },
     {
       "seconds": 0.9811290180000469,
       "bytes": 1073130,
@@ -562,6 +556,12 @@
       "bytes": 6956463,
       "pages": 35,
       "shards": 2
+    },
+    {
+      "seconds": 1.286864922998575,
+      "bytes": 7382695,
+      "pages": 34,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790434090210070261-a05fc3fb/train.jsonl

@@ -102,3 +102,4 @@
 {"step": 50200, "loss": 0.6411357149481773, "lr": 0.0003, "grad_norm": 0.1354083447785202, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 7346867699, "tokens_per_second": 147604.6036964836, "data_wait_s": 1.146300019172486e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 50300, "loss": 0.6121636796742678, "lr": 0.0003, "grad_norm": 0.12666534850508873, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 7361338036, "tokens_per_second": 148150.24210174807, "data_wait_s": 1.0711000868468545e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 50400, "loss": 0.6780245490372181, "lr": 0.0003, "grad_norm": 0.19957534603921212, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 7375921768, "tokens_per_second": 148934.62356921754, "data_wait_s": 8.644999979878776e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 50500, "loss": 0.5936776921153069, "lr": 0.0003, "grad_norm": 0.12213298037925359, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 7390598106, "tokens_per_second": 150642.64040251233, "data_wait_s": 1.0911999197560363e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790434090.372bd512-57cc-4928-b44a-3b375a0fa071-5spg5.451.0.1790434090210070261-a05fc3fb

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:2a96767f868fa4e922379f5cbfdadbd27be57404ec91a57bae4c680acc58a477
-size 222003
+oid sha256:68221c9265244b4227141275a94a934aeba524ba5b47d99c37a6e17718dcf259
+size 224121